CN110458340B - Autoregressive prediction method of building air conditioning cooling load based on pattern classification - Google Patents

Abstract

The invention discloses an auto-regression prediction method for building air-conditioning cooling load based on pattern classification. It can be applied to scientific research and engineering applications in related fields of building air conditioning systems. The method includes: preprocessing the input data of all models to obtain the original data set; performing cluster analysis on the air-conditioning load, that is, pre-classifying the load pattern; determining the influencing factors of the two time scale prediction models of the day before and the day, and clearly The input selection of the day-ahead and day-ahead forecasting model; using the Spearman correlation coefficient analysis method to check the correlation between the cooling load and the influencing factors; determining the load pattern of the forecast day; finally, establishing the day-ahead and day cooling load forecasting model. The prediction method can improve the accuracy of the prediction results of the building air-conditioning cooling load, accelerate the calculation speed of the prediction model, and guide the operation and regulation of the building air-conditioning system.

Description

Autoregressive prediction method for building air conditioning cooling load based on pattern classification

Technical Field

The invention relates to a building air-conditioning cooling load prediction method, in particular to a building air-conditioning cooling load autoregression prediction method based on pattern classification.

Background Art

Many studies have shown that the energy consumed by heating, ventilation and air conditioning (HVAC) systems is a major component of total building energy consumption. Most current air conditioning systems use traditional feedback control methods to adjust the system based on the return water temperature on the user side. However, due to the complex structure of buildings and the changeable behavior of personnel, feedback control can no longer meet people's needs. By predicting the cooling load of buildings in the future, it is possible to respond to changes in indoor conditions in real time. Therefore, cooling load prediction is an effective way to improve the energy efficiency of air conditioning systems. Cooling load prediction can notify operators of future cooling needs in advance. Operators can manage and set the system based on the predicted cooling load, and the regulation of the HVAC system changes from "feedback" mode to "feedforward" mode. The prediction models used to guide system management and operation can be roughly divided into day-ahead and day-ahead prediction models. Day-ahead prediction models are used to determine the cooling load for the next day. They can guide operators to prepare HVAC system management plans in advance, such as arranging chillers, pumps used on the forecast day, cooling towers, etc., and maintaining or repairing other equipment. The day-ahead prediction model is used to determine the cooling load for the forecast time in the next few hours. It can accurately predict the cooling demand over a period of time and guide operators to formulate HVAC system operation strategies.

The autoregressive model uses historical data from previous periods of the same variable to simulate and predict future data, and assumes that they are in a linear relationship, which includes the lag term of the endogenous variable. The ARX model is an autoregressive model with external input. This model adds external input on the basis of this autoregressive model, which is more in line with the data characteristics of the cooling load. In previous studies, the ARX model is often used to directly predict the cooling load of a building. However, the historical data and related parameters required to predict the cooling load at different times are different, and the calculation results of the ARX model will also have large errors.

Therefore, adopting a reasonable method to establish an autoregressive prediction method for building air conditioning cooling load based on pattern classification is a key issue that needs to be solved urgently to reasonably manage and set up the system and reduce system energy consumption.

Summary of the invention

In view of this, the present invention provides a building air conditioning cooling load autoregressive prediction method based on pattern classification to solve the above problems.

Based on previous inventions, the present invention has made the following improvements: using the interquartile range method to preprocess the data, detecting outliers in the original building data, and obtaining the original data set; using k-means cluster analysis to perform load pattern cluster analysis to obtain a typical day of load pattern; determining the factors affecting the day-ahead prediction and the day-of-the-day prediction, using the Spearman coefficient analysis method to perform a correlation analysis on the building air-conditioning cooling load and the influencing factors, eliminating variables with low correlation, and obtaining the model input selection of the day-ahead and day-of-the-day prediction models; using k-means cluster analysis to determine the load pattern of the prediction day; using an autoregressive model based on pattern classification to establish a day-ahead and day-of-the-day cooling load prediction model.

The present invention proposes a building air conditioning cooling load autoregressive prediction method based on pattern classification, comprising the following steps:

The interquartile range method is used to preprocess the data, detect outliers in the original building data, and obtain the original data set;

K-means cluster analysis was used to classify the building cooling load, and load pattern cluster analysis was performed to obtain the typical day of the load pattern;

Determine the factors that affect the day-ahead forecast and the same-day forecast, conduct correlation analysis on the building air conditioning cooling load and the influencing factors, eliminate the variables with low correlation, and obtain the model input selection for the day-ahead and same-day forecast models;

The load pattern of the forecast day was determined using k-means cluster analysis;

The autoregressive model based on pattern classification is used to establish the cooling load forecasting model for the day before and the day before.

Evaluate building air conditioning cooling load prediction models.

Furthermore, the influencing factors include factors influencing the cooling load prediction of the day before and factors influencing the cooling load prediction of the day.

Furthermore, all data were preprocessed using the interquartile range method.

Furthermore, the load pattern clustering analysis of the building cooling load is carried out, and the method adopted is the k-means clustering method. First, k objects are randomly selected as the initial cluster centers; second, the distance between each object and each sub-cluster center is calculated, and each object is assigned to the cluster center closest to it; then, a new cluster center is calculated; finally, the above two steps are repeated until the error square sum is locally minimized.

The formula for calculating the sum of squared errors is

Where: SSE represents the sum of squared errors;

_Ci represents the cluster center of cluster _Si ;

K represents the number of cluster centers;

x represents a data point in cluster _Si .

The load day closest to the cluster center of each cluster will be considered as a typical day.

Furthermore, the correlation analysis between the building air conditioning cooling load and influencing factors was conducted, and the Spearman coefficient was used to calculate the size and direction of the correlation between various variables.

The method to calculate the Spearman coefficient is

Where: N represents the number of data;

d _i represents the difference in the ranking of two variables in the i-th data.

The following selection criteria are followed when selecting the influencing factors. The correlation coefficient of the influencing factors is compared with the specified limit value. When the limit value is met, it is considered to have a significant impact on the predicted load and is then extracted as the model input. There is no unified method to select the limit value; the user should reasonably select the limit value based on the application. However, the general requirement for the limit value is greater than 0.5, which indicates that there is at least a relationship between them.

represents a sample of the variable; represents a sample of the variable; represents the number of samples of the variable and. Further, k-means cluster analysis is used to determine the load pattern of the forecast day.

Furthermore, an autoregressive model based on pattern classification is used to establish a cooling load forecasting model for the day before and the day before.

The autoregressive model based on pattern classification (ARX model) is used to describe the relationship between the predicted cooling load and various model inputs. The selected model input is the parameter with the strongest correlation with the cooling load obtained by the Spearman coefficient method.

Beneficial effects: This prediction method can improve the accuracy of building air-conditioning cooling load prediction results, speed up the calculation of the prediction model, and guide the operation and regulation of the building air-conditioning system.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a flow chart of the present invention;

Figure 2 shows the load pattern clustering results.

DETAILED DESCRIPTION

In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings. As shown in FIG1 , this embodiment provides a cooling load model prediction method based on an autoregressive model with additional input and an artificial neural network model, comprising the following steps:

Step 1: Use the interquartile range method to preprocess the data, detect outliers in the original building data, and obtain the original data set;

The air conditioning system of a certain building runs 24 hours a day. The air conditioning system of the building consists of 8 chillers of the same specifications. The rated cooling capacity of one chiller is 4186 kilowatts. The cooling tower, cooling water pump and primary side chilled water pump in the air conditioning system are all matched with the chillers one by one. The secondary side chilled water pump is adjusted according to the air conditioning terminal. The chilled water outlet temperature and water volume are both rated design values. The actual cooling capacity data of the building is measured by the heat meter, and the outdoor dry bulb temperature data is measured by the meteorological station, and the data recording interval is 1 hour. Based on the cooling capacity data records of the building in the two cooling seasons of 2017 and 2018, the specific implementation mode of the present invention is described in detail.

Step 2: Use k-means clustering method to classify the building cooling load, perform load pattern clustering analysis, and obtain the typical day of load pattern;

First, randomly select k objects as the initial cluster centers; second, calculate the distance between each object and each sub-cluster center, and assign each object to the cluster center closest to it; then, calculate the new cluster center; finally, repeat the above two steps until the error sum of squares is locally minimized. The calculation formula of the error sum of squares is shown in formula (1).

Where: SSE represents the sum of squared errors;

_Ci represents the cluster center of cluster _Si ;

K represents the number of cluster centers;

x represents a data point in cluster _Si ; the load day closest to the center of each cluster will be considered a typical day.

The hourly cooling loads in 2017 and 2018 were clustered based on the K-means clustering method. The number of clusters was determined by the sum of squared errors, and the results are shown in Figure 2. When the number of clusters increased to 5, the reduction of SSE slowed down. Therefore, five load patterns were set in the two-year cooling season.

Step 3: Determine the factors that affect the day-ahead forecast and the same-day forecast, conduct a correlation analysis between the building air conditioning cooling load and the influencing factors, eliminate the variables with low correlation, and obtain the model input selection for the day-ahead and same-day forecast models;

The correlation analysis of building air conditioning cooling load and influencing factors is carried out, and the Spearman coefficient is used to calculate the size and direction of the correlation of various variables. The method for calculating the Spearman coefficient is shown in formula (2).

Where: N represents the number of data;

d _i represents the difference in the ranking of two variables in the i-th data.

The following selection criteria are followed when selecting the influencing factors. The correlation coefficient of the influencing factor is compared with the specified limit value. When the limit value is met, it is considered to have a significant impact on the predicted load and is then extracted as the model input. There is no unified method to select the limit value; the limit value should be reasonably selected based on the application. However, the limit value is generally required to be greater than 0.5, which indicates that there is at least a relationship between them.

Under the same load pattern, the relationship between the historical data load and the predicted day load is determined by correlation analysis. In the case building, taking cluster 5 as an example, when 0.4 is used as the limit value of the correlation coefficient, it is found that the loads at t-24 and t-48 have the strongest correlation with the predicted load. Therefore, the loads at T-24 and T-48 are used as the model input for the prediction model for cluster 5 a day ago. For clusters 2 to 4, the same analysis is performed to obtain the model input. For cluster 1, which has almost no cooling demand throughout the day, when the forecast day is determined to be cluster 1, the hourly load on that day is considered to be zero, so no correlation analysis is performed.

represents a sample of the variable; represents a sample of the variable; represents the number of samples of the variable and. Due to the thermal inertia of the building envelope, there will be a certain lag in the trend of changes in the outdoor dry-bulb temperature transmitted to the interior. The outdoor dry-bulb temperature in the historical period also affects the predicted load. Through correlation analysis, the relationship between the cooling load at the prediction time t and the outdoor dry-bulb temperature at the prediction time and the historical time is determined. The impact of the outdoor dry-bulb temperature on the predicted load is the same in all load modes. Therefore, using the data of the entire cooling season, the relationship between the predicted load and the outdoor dry-bulb temperature is analyzed.

Step 4 uses k-means cluster analysis to determine the load pattern of the forecast day;

First, randomly select k objects as the initial cluster centers; second, calculate the distance between each object and each sub-cluster center, and assign each object to the cluster center closest to it; then, calculate the new cluster center; finally, repeat the above two steps until the error sum of squares is locally minimized. The calculation formula of the error sum of squares is shown in formula (3).

Where: SSE represents the sum of squared errors;

_Ci represents the cluster center of cluster _Si ;

K represents the number of cluster centers;

x represents a data point in cluster _Si .

All load days in the cooling seasons of 2017 and 2018 were randomly divided into 10 data sets. The load days in each data set are distributed in all stages of the entire cooling season, and each data set contains five load patterns, so the model validation is universal. The classification results are shown in Table 1.

Table 1 k-means clustering analysis results

Table 1 shows that the SSE accuracy ranges from 0.5 to 0.9, and the average classification accuracy is about 0.7. From all datasets, the dataset with higher classification accuracy (dataset 7) and the dataset with lower classification accuracy (dataset 9) are selected for subsequent analysis.

Step 5: Use the autoregressive model with additional input to establish the cooling load forecasting model for the day before and the day before;

The autoregressive model with additional inputs (ARX model) is used to describe the relationship between the predicted cooling load and various model inputs. In this case, the selected input parameters are outdoor dry-bulb temperature and historical cooling load. The method used is shown in Equation (4).

Where: Q _t represents the cooling load at the predicted time t;

_Ti represents the outdoor dry-bulb temperature;

_Qi represents the historical cooling load;

m represents the number of outdoor dry-bulb temperature values that affect Q _t ;

_wi represents the coefficient of the model input

w _l represents a constant offset factor, which is used to partially reduce the impact of modeling errors.

The ARX prediction technology is used to establish the day-ahead prediction model for data clusters 7 and 9. Taking data cluster 7 as an example, Table 2 gives the ARX prediction model for clusters 2 to 5.

Table 2 ARX prediction models for clusters 2 to 5

Although the present invention has been described above in conjunction with the accompanying drawings, the present invention is not limited to the above-mentioned specific embodiments, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can make many modifications without departing from the purpose of the present invention, which are all within the protection of the present invention.

Claims (2)

1. The autoregressive prediction method for the cold load of the building air conditioner based on the pattern classification is characterized by comprising the following steps of:

(1) Preprocessing data by adopting a quarter bit distance method, and detecting abnormal values in building original data;

(2) Classifying the building air conditioner cold load obtained in the step (1) by adopting k-means cluster analysis, and carrying out cluster analysis of load modes to obtain a load mode of a typical day;

(3) Determining factors influencing the day-ahead prediction and the day-ahead prediction, performing correlation analysis on the building air conditioner cold load and the influencing factors, removing variables with lower correlation, and obtaining model input choices of the day-ahead and the day-ahead prediction models;

(4) Determining a load mode of a predicted day by adopting k-means cluster analysis;

(5) According to the load mode obtained in the step (4), an autoregressive model is adopted to establish a prediction model of the air conditioner cold load before and at the same time;

the influence factors comprise a prediction factor for influencing the daily cold load and a prediction factor for influencing the daily cold load;

the method adopted in the step (2) for carrying out load pattern cluster analysis on the building air conditioner load and determining the load pattern of the predicted day in the step (4) is k-means cluster analysis:

firstly, randomly selecting k objects as initial clustering centers;

secondly, calculating the distance between each object and each sub-cluster center, and distributing each object to the cluster center closest to the object;

then, calculating a new cluster center;

finally, repeating the two steps until the square sum of errors is locally minimum, wherein the calculation formula of the square sum of errors is as follows:

wherein: SSE represents the sum of squares of errors;

C _i representing clusters S _i Is a cluster center of the group (C);

k represents the number of clustering centers;

x represents cluster S _i Data points in (a);

the load day closest to each cluster center will be considered the typical day;

in the step (3), correlation analysis is carried out on the cold load of the building air conditioner and influence factors, and the magnitude and the direction of correlation of various variables are calculated by adopting a spearman correlation coefficient:

wherein: n represents the number of data;

d _i representing the difference in ordering of the two variables in the ith data.

2. The method for autoregressive prediction of building air conditioner cold load based on pattern classification according to claim 1, wherein in the step (5), an autoregressive model based on pattern classification is used to build a model for predicting the building air conditioner cold load before and on the same day.

Images